Erik Anderson & Tyler Cruickshank
Maximizing agricultural productivity is critical for maintaining world food supply and controlling food cost. Modern agribusiness employs targeted technology to optimize cropland productivity using modern information and communication technologies at fine spatial resolution (within-field scale).
Hyrdrosat is a value-added satellite data provider whose mission is to provide real-time thermal infrared data to customers in the agribusiness and government sectors. Thermal infrared data can be used to inform adaptive irrigation and predict crop yields .
Until the upcoming launch of its own 16-satellite constellation, Hydrosat produces estimates of land surface temperature using thermal infrared data from a combination of MODIS, Sentinel and Landsat imaging platforms. Available data suffers from low spatial and temporal resolutions. Hydrosat utilizes an implementation of the data mining approach for sharpening thermal satellite imagery (DMS; as described in Gao 2012) and a separate algorithm for interpolating land surface temperature between acquisition windows (STARFM; as described in Gao 2006), to produce a proprietry "Fused" Land Surface Temperature (LST) imagery product resulting in a near daily estimate of LST at sub-30 meter resolution.
The Fused LST product and its inputs are accessed from a Spatial Temporal Asset Catalog (STAC). STAC is a specification standard with a unified language to describe geospatial data which allows it to be more easily searchable and queryable. Data are served through the Hydrosat's Fusion Hub. (Credentials for data access must be requested as directed on the Fusion Hub homepage.)
This project aims to explore the input data and intermediate outputs of the algorithms used to produce the LST product in order to detect errors potentially introduced by data quality isses or the application of the algorithms while also exploring applications within agricultural contexts.
Data range from 2016-08-13 07:00:00+00:00 to 2023-01-01 06:30:00+00:00
The project is focused on agricultural areas with active farming in the San Joaquin Valley of California, United States.
Figure 1 | This map visualizes the project area for this workflow. The red boundary illustrates the clip area for imagery. The black polygon displays the field boundaries for the agricultural field evaluated. The black marker shows the field center and is the location from which imagery will be sampled when calculating key metrics. The green marker shows the location of the meteorological tower.
Our work revealed two important findings for Hydrosat's Fused Land Surface Temperature Product:
Each finding is discussed in detail below.
When compositing Sentinel-2 and Landsat imagery, the current implementation of the STARFM algorithm to interpolate between acquisition windows tends to suppress the vegetation signal. In the current implementation, the median pixel value is calculated for the period two weeks before and two weeks after the interpolated value. This moving window approach reduces the sensitivity of the resulting data to crop stress in the short term.
To illustrate, we compare the Normalized Difference Vegetation Index (NDVI) as calculated by the meterological tower, Senitnel 2 data, and the composites for Sentinel 2 and MODIS. NDVI is a commonly used remotely sensed index that provides data on the "greenness" or health of crop vegetation.
found 30 items
c:\Users\Erik_Anderson\miniconda3\envs\ea-lst\Lib\site-packages\rasterio\warp.py:344: NotGeoreferencedWarning: Dataset has no geotransform, gcps, or rpcs. The identity matrix will be returned. _reproject(
Figure 2 | NDVI is plotted over the growing season for the year 2022 in this alfalfa field. The in-situ meteorological tower provides a benchmark for expected NDVI values. Both Sentinel-2 and the MODIS composite NDVI values are consistent with the periodicity of the benchmark. However, the Sentinel-2 composite reflects only the general trend towards lower NDVI across the entire growing season and does not capture the periodicity of the benchmark.
We can observe this effect more directly by comparing the true color imagery from Sentinel-2 and the composite Sentinel-2 data. The first panel shows the original Sentinel-2 true color imagery and the second panel shows the composite Sentinel-2 imagery.
Figure 3 Panel A | A time series of Sentinel-2 true color imagery for the growing season in the year 2022. The imagery illustrates the periodic greening and harvest of an alfalfa field throughout the growing season.
Figure 3 Panel B | A time series of Sentinel-2 true color imagery for the growing season in the year 2022. The imagery illustrates the muted vegetation signal caused by compositing the imagery with a moving window approach. The periodic harvest and intermittent dry up of the alfalfa field is hidden.
In the original Sentinel-2 true color imagery, we can observe a periodic greening and drying up of the field. In the Central Valley of California, alfalfa is often harvested across many cuttings each year. The imagery is consistent with this practice and reflects the NDVI measures produced by the meteorological tower on site.
However, the composite Sentinel-2 data shows a more consistent trend from green vegetation to dry soil over the course of the growing season. While this is the general trend for this field, the composite Sentinel-2 data does not reflect the periodic nature of this trend and thus will likely miss important cues of crop stress.
NDVI index is a lagging indicator of crop stress because water or heat stress occurs prior to the degradation of vegetation greenness.
[Include NDVI box plots vs CATD]
Gao, F. Masek, J. Schwaller, M. Hall, F. 2006. On the Blending of Landsat and MODIS Surface Reflectance: Predicting Landsat Surface Reflectance. IEE Transaction on Geoscience and Remote Sensing Vol 44, No 8.
Gao, F. Kustas, W. Anderson, M. 2012. A Data Mining Approach for Sharpening Thermal Satellite Imagery Over Land. Remote Sensing. doi: 10.3390/rs4113287.
Karnieli, A., Agam, N., Pinker, R., Anderseon, M., Imhoff, M., Gutman, G., Panov, N., Goldberg, A., Use of NDVI and Land Surface Temperature for Drought Assessment: Merits and Limitations. Journal of Climate. Vol 23. 2010
Y. Erdem , T. Erdem , A.H. Orta & H. Okursoy. Canopy-air temperature differential for potato under different irrigation regimes. Acta Agriculturae Scandinavica Section B-Soil and Plant Science, 2006; 56: 206216